What Anthropic Released
On June 30, 2026, Anthropic introduced Claude Sonnet 5. This isn't a minor update: it's the most capable and most agentic model in the Sonnet tier ever built, already available on every plan — it's the default model for Free and Pro, and it's accessible to Max, Team, and Enterprise.
The promise, in one line: performance close to Opus 4.8, the flagship model, at a much lower cost. For anyone using AI intensively, that's the headline that matters most, because it shifts the cost/quality equation.
The Numbers: How Close It Gets to Opus
The leap shows up on the terrain that carries the most weight today — agentic work: planning, using tools like a browser and a terminal, running tasks autonomously.
On SWE-bench Pro — the agentic coding benchmark — Sonnet 5 scores 63.2%, versus 58.1% for the previous Sonnet 4.6 and 69.2% for Opus 4.8. In other words: Sonnet 5 closes most of the gap with the flagship model while staying in the Sonnet price tier. It's also safer than 4.6 in agentic contexts, which is no small thing for anyone putting agents into production.
The Price Is the Real Story
This is where the news lands for businesses. Sonnet 5 launches with introductory pricing: $2 per million input tokens and $10 per million output tokens, through August 31, 2026. After that, it moves to $3 / $15.
That's a mid-tier price for quality approaching the top of the range. For high-volume use cases — agents running continuously, automations, large-scale document processing — it means drastically lowering the cost per unit of value. And the introductory window through August is effectively an invitation to test it on real volumes now. For the full breakdown, see how much Claude costs for businesses.
Want to use Sonnet 5 to cut your AI costs?
30 minutes to discuss your specific case.
When You Still Need Opus
Sonnet 5 doesn't make Opus useless: it makes it a more deliberate choice.
Opus 4.8 stays ahead on tasks that demand maximum reasoning and depth — the most complex analyses, long multi-step reasoning, the cases where that ~6% benchmark gap translates into a better result worth the higher cost. The right logic is routing: Sonnet 5 as the workhorse for the bulk of tasks and for agents at volume, Opus for the critical jobs where quality isn't negotiable. To orient yourself, see Sonnet vs Opus vs Haiku.
The Context: The Race for Cheap Agentic Models
Sonnet 5 doesn't launch in a vacuum. In the same days, OpenAI introduced GPT-5.6 with its Sol/Terra/Luna scale, also aimed at agents and cost. The market's message is clear: the frontier is no longer just "the most powerful model," but "the highest quality at the lowest cost to run agents at scale."
For businesses, that's good news: competitive pressure is pushing prices down and capabilities up at the same time. Those who know how to choose and route the right models for the right tasks come out ahead.
What to Do Now, in Practice
Three concrete moves. First: if you already use Claude, consider shifting the workloads you currently run on Sonnet 4.6 — or on Opus out of pure habit — over to Sonnet 5, because you're probably paying less for equal (or better) results. Second: use the introductory window through August 31 to test it on real volumes and measure cost per result. Third: revisit your routing logic — which task goes to which model — because with Sonnet 5 the balance has shifted.
If you want to figure out how to reorganize your use of Claude models in your context and how much you can save, let's talk.